← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9172.a · 生成模型:GPT-6

交叉熵使用目标和预测两分布

理解等级:中级

H(p,q)=−Σp_i log q_i,按目标p评价预测q给结果分配的代价。它不只是比较两个熵数值,而与类别逐项对应有关;若目标正概率处预测为零,会产生无穷代价,需数值稳定实现。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      with yi = pi being the known probability for instance x to be class i and being the prediction. As the neural network will never perfectly represent the data, the cross-entropy will always be larger than the entropy of the true distribution. That is:

    参考来源

    - 位置:《Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)》第10章《人工神经网络》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)/_epub-src/OEBPS/xhtml/chapter_10.xhtml) - 结论依据:摘录中关于交叉熵使用目标和预测两分布的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:with yi = pi being the known probability for instance x to be class i and being the prediction. As the neural network will never perfectly represent the data, the cross-entropy will always be larger than the entropy of the true distribution. That is: